Applying reinforcement learning to mitigate wake-induced lift fluctuation of a wall-confined circular cylinder in tandem configuration.

The flow around two tandem circular cylinders leads to significant lift fluctuation in the downstream cylinder owing to periodic vortex shedding. To address such research issues, we present herein a numerical study that uses deep reinforcement learning to perform active flow control (AFC) on two tandem cylinders with a low Reynolds number of 100, where the actuator causes the rotation of the downstream cylinder. First, the cylinder center spacing ratio L * varies from 1.5 to 9.0, and the variation of L * leads to the quasi-steady reattachment regime ( L * ≤ 3.5) and the co-shedding regime ( L * ≥ 4.0). The fluctuating lift of the downstream cylinder is maximum when L * = 4.5. Next, we train an optimal AFC strategy that suppresses 75% of the lift fluctuation in the downstream cylinder. This approach differs from using direct-opposition control to change the vortex-shedding frequency or strength, as reported in previous studies. This strategy modifies the phase difference between the lift fluctuations of the two cylinders by delaying the merging with the upstream cylinder wake and accelerating the formation of recirculating bubbles after the vortex merging. With the new phase difference, the effect of the additional lift from the upstream cylinder is significantly mitigated. The results of the dynamic mode decomposition show that the vortices surrounding the downstream cylinder in mode 1 that contribute to the lift fluctuation are weakened. To the best of our knowledge, this investigation can provide new ideas and physical insights into the problem of AFC under disturbed incoming flow. [ABSTRACT FROM AUTHOR]